bcache: Write out full stripes

Now that we're tracking dirty data per stripe, we can add two
optimizations for raid5/6:

 * If a stripe is already dirty, force writes to that stripe to
   writeback mode - to help build up full stripes of dirty data

 * When flushing dirty data, preferentially write out full stripes first
   if there are any.

Signed-off-by: Kent Overstreet <koverstreet@google.com>

9 files changed, 121 insertions, 37 deletions

diff --git a/drivers/md/bcache/bcache.h b/drivers/md/bcache/bcache.h
index dbddef0cdb5..342ba86c6e4 100644
--- a/drivers/md/bcache/bcache.h
+++ b/drivers/md/bcache/bcache.h
@@ -387,8 +387,6 @@ struct keybuf_key {
 typedef bool (keybuf_pred_fn)(struct keybuf *, struct bkey *);
 
 struct keybuf {
-	keybuf_pred_fn		*key_predicate;
-
 	struct bkey		last_scanned;
 	spinlock_t		lock;
 
@@ -532,6 +530,7 @@ struct cached_dev {
 	unsigned		sequential_merge:1;
 	unsigned		verify:1;
 
+	unsigned		partial_stripes_expensive:1;
 	unsigned		writeback_metadata:1;
 	unsigned		writeback_running:1;
 	unsigned char		writeback_percent;
diff --git a/drivers/md/bcache/btree.c b/drivers/md/bcache/btree.c
index b93cf56260a..09fb8a2f43d 100644
--- a/drivers/md/bcache/btree.c
+++ b/drivers/md/bcache/btree.c
@@ -2252,7 +2252,8 @@ static inline int keybuf_nonoverlapping_cmp(struct keybuf_key *l,
 }
 
 static int bch_btree_refill_keybuf(struct btree *b, struct btree_op *op,
-				   struct keybuf *buf, struct bkey *end)
+				   struct keybuf *buf, struct bkey *end,
+				   keybuf_pred_fn *pred)
 {
 	struct btree_iter iter;
 	bch_btree_iter_init(b, &iter, &buf->last_scanned);
@@ -2271,7 +2272,7 @@ static int bch_btree_refill_keybuf(struct btree *b, struct btree_op *op,
 			if (bkey_cmp(&buf->last_scanned, end) >= 0)
 				break;
 
-			if (buf->key_predicate(buf, k)) {
+			if (pred(buf, k)) {
 				struct keybuf_key *w;
 
 				spin_lock(&buf->lock);
@@ -2290,7 +2291,7 @@ static int bch_btree_refill_keybuf(struct btree *b, struct btree_op *op,
 			if (!k)
 				break;
 
-			btree(refill_keybuf, k, b, op, buf, end);
+			btree(refill_keybuf, k, b, op, buf, end, pred);
 			/*
 			 * Might get an error here, but can't really do anything
 			 * and it'll get logged elsewhere. Just read what we
@@ -2308,7 +2309,7 @@ static int bch_btree_refill_keybuf(struct btree *b, struct btree_op *op,
 }
 
 void bch_refill_keybuf(struct cache_set *c, struct keybuf *buf,
-			  struct bkey *end)
+		       struct bkey *end, keybuf_pred_fn *pred)
 {
 	struct bkey start = buf->last_scanned;
 	struct btree_op op;
@@ -2316,7 +2317,7 @@ void bch_refill_keybuf(struct cache_set *c, struct keybuf *buf,
 
 	cond_resched();
 
-	btree_root(refill_keybuf, c, &op, buf, end);
+	btree_root(refill_keybuf, c, &op, buf, end, pred);
 	closure_sync(&op.cl);
 
 	pr_debug("found %s keys from %llu:%llu to %llu:%llu",
@@ -2402,7 +2403,8 @@ struct keybuf_key *bch_keybuf_next(struct keybuf *buf)
 
 struct keybuf_key *bch_keybuf_next_rescan(struct cache_set *c,
 					     struct keybuf *buf,
-					     struct bkey *end)
+					     struct bkey *end,
+					     keybuf_pred_fn *pred)
 {
 	struct keybuf_key *ret;
 
@@ -2416,15 +2418,14 @@ struct keybuf_key *bch_keybuf_next_rescan(struct cache_set *c,
 			break;
 		}
 
-		bch_refill_keybuf(c, buf, end);
+		bch_refill_keybuf(c, buf, end, pred);
 	}
 
 	return ret;
 }
 
-void bch_keybuf_init(struct keybuf *buf, keybuf_pred_fn *fn)
+void bch_keybuf_init(struct keybuf *buf)
 {
-	buf->key_predicate	= fn;
 	buf->last_scanned	= MAX_KEY;
 	buf->keys		= RB_ROOT;
 
diff --git a/drivers/md/bcache/btree.h b/drivers/md/bcache/btree.h
index 2b016b93cad..f66d69a7baf 100644
--- a/drivers/md/bcache/btree.h
+++ b/drivers/md/bcache/btree.h
@@ -391,13 +391,14 @@ void bch_moving_gc(struct closure *);
 int bch_btree_check(struct cache_set *, struct btree_op *);
 uint8_t __bch_btree_mark_key(struct cache_set *, int, struct bkey *);
 
-void bch_keybuf_init(struct keybuf *, keybuf_pred_fn *);
-void bch_refill_keybuf(struct cache_set *, struct keybuf *, struct bkey *);
+void bch_keybuf_init(struct keybuf *);
+void bch_refill_keybuf(struct cache_set *, struct keybuf *, struct bkey *,
+		       keybuf_pred_fn *);
 bool bch_keybuf_check_overlapping(struct keybuf *, struct bkey *,
 				  struct bkey *);
 void bch_keybuf_del(struct keybuf *, struct keybuf_key *);
 struct keybuf_key *bch_keybuf_next(struct keybuf *);
-struct keybuf_key *bch_keybuf_next_rescan(struct cache_set *,
-					  struct keybuf *, struct bkey *);
+struct keybuf_key *bch_keybuf_next_rescan(struct cache_set *, struct keybuf *,
+					  struct bkey *, keybuf_pred_fn *);
 
 #endif
diff --git a/drivers/md/bcache/debug.c b/drivers/md/bcache/debug.c
index 82e3a07771e..1c8fd319846 100644
--- a/drivers/md/bcache/debug.c
+++ b/drivers/md/bcache/debug.c
@@ -357,7 +357,7 @@ static ssize_t bch_dump_read(struct file *file, char __user *buf,
 		if (i->bytes)
 			break;
 
-		w = bch_keybuf_next_rescan(i->c, &i->keys, &MAX_KEY);
+		w = bch_keybuf_next_rescan(i->c, &i->keys, &MAX_KEY, dump_pred);
 		if (!w)
 			break;
 
@@ -380,7 +380,7 @@ static int bch_dump_open(struct inode *inode, struct file *file)
 
 	file->private_data = i;
 	i->c = c;
-	bch_keybuf_init(&i->keys, dump_pred);
+	bch_keybuf_init(&i->keys);
 	i->keys.last_scanned = KEY(0, 0, 0);
 
 	return 0;
diff --git a/drivers/md/bcache/movinggc.c b/drivers/md/bcache/movinggc.c
index 04f6b97ffda..a241e9fd4f7 100644
--- a/drivers/md/bcache/movinggc.c
+++ b/drivers/md/bcache/movinggc.c
@@ -136,7 +136,8 @@ static void read_moving(struct closure *cl)
 	/* XXX: if we error, background writeback could stall indefinitely */
 
 	while (!test_bit(CACHE_SET_STOPPING, &c->flags)) {
-		w = bch_keybuf_next_rescan(c, &c->moving_gc_keys, &MAX_KEY);
+		w = bch_keybuf_next_rescan(c, &c->moving_gc_keys,
+					   &MAX_KEY, moving_pred);
 		if (!w)
 			break;
 
@@ -248,5 +249,5 @@ void bch_moving_gc(struct closure *cl)
 
 void bch_moving_init_cache_set(struct cache_set *c)
 {
-	bch_keybuf_init(&c->moving_gc_keys, moving_pred);
+	bch_keybuf_init(&c->moving_gc_keys);
 }
diff --git a/drivers/md/bcache/request.c b/drivers/md/bcache/request.c
index 017c95fced8..17bd59704eb 100644
--- a/drivers/md/bcache/request.c
+++ b/drivers/md/bcache/request.c
@@ -22,8 +22,6 @@
 
 #define CUTOFF_CACHE_ADD	95
 #define CUTOFF_CACHE_READA	90
-#define CUTOFF_WRITEBACK	50
-#define CUTOFF_WRITEBACK_SYNC	75
 
 struct kmem_cache *bch_search_cache;
 
@@ -998,17 +996,6 @@ static void cached_dev_write_complete(struct closure *cl)
 	cached_dev_bio_complete(cl);
 }
 
-static bool should_writeback(struct cached_dev *dc, struct bio *bio)
-{
-	unsigned threshold = (bio->bi_rw & REQ_SYNC)
-		? CUTOFF_WRITEBACK_SYNC
-		: CUTOFF_WRITEBACK;
-
-	return !atomic_read(&dc->disk.detaching) &&
-		cache_mode(dc, bio) == CACHE_MODE_WRITEBACK &&
-		dc->disk.c->gc_stats.in_use < threshold;
-}
-
 static void request_write(struct cached_dev *dc, struct search *s)
 {
 	struct closure *cl = &s->cl;
@@ -1030,12 +1017,16 @@ static void request_write(struct cached_dev *dc, struct search *s)
 	if (bio->bi_rw & REQ_DISCARD)
 		goto skip;
 
+	if (should_writeback(dc, s->orig_bio,
+			     cache_mode(dc, bio),
+			     s->op.skip)) {
+		s->op.skip = false;
+		s->writeback = true;
+	}
+
 	if (s->op.skip)
 		goto skip;
 
-	if (should_writeback(dc, s->orig_bio))
-		s->writeback = true;
-
 	trace_bcache_write(s->orig_bio, s->writeback, s->op.skip);
 
 	if (!s->writeback) {
diff --git a/drivers/md/bcache/sysfs.c b/drivers/md/bcache/sysfs.c
index cf8d91ec323..70c6dff0d0c 100644
--- a/drivers/md/bcache/sysfs.c
+++ b/drivers/md/bcache/sysfs.c
@@ -81,6 +81,9 @@ rw_attribute(writeback_rate_p_term_inverse);
 rw_attribute(writeback_rate_d_smooth);
 read_attribute(writeback_rate_debug);
 
+read_attribute(stripe_size);
+read_attribute(partial_stripes_expensive);
+
 rw_attribute(synchronous);
 rw_attribute(journal_delay_ms);
 rw_attribute(discard);
@@ -147,6 +150,9 @@ SHOW(__bch_cached_dev)
 	sysfs_hprint(dirty_data,
 		     bcache_dev_sectors_dirty(&dc->disk) << 9);
 
+	sysfs_hprint(stripe_size,	(1 << dc->disk.stripe_size_bits) << 9);
+	var_printf(partial_stripes_expensive,	"%u");
+
 	var_printf(sequential_merge,	"%i");
 	var_hprint(sequential_cutoff);
 	var_hprint(readahead);
@@ -286,6 +292,8 @@ static struct attribute *bch_cached_dev_files[] = {
 	&sysfs_writeback_rate_d_smooth,
 	&sysfs_writeback_rate_debug,
 	&sysfs_dirty_data,
+	&sysfs_stripe_size,
+	&sysfs_partial_stripes_expensive,
 	&sysfs_sequential_cutoff,
 	&sysfs_sequential_merge,
 	&sysfs_clear_stats,
diff --git a/drivers/md/bcache/writeback.c b/drivers/md/bcache/writeback.c
index dd815475c52..d81ee5ccc72 100644
--- a/drivers/md/bcache/writeback.c
+++ b/drivers/md/bcache/writeback.c
@@ -108,6 +108,31 @@ static bool dirty_pred(struct keybuf *buf, struct bkey *k)
 	return KEY_DIRTY(k);
 }
 
+static bool dirty_full_stripe_pred(struct keybuf *buf, struct bkey *k)
+{
+	uint64_t stripe;
+	unsigned nr_sectors = KEY_SIZE(k);
+	struct cached_dev *dc = container_of(buf, struct cached_dev,
+					     writeback_keys);
+	unsigned stripe_size = 1 << dc->disk.stripe_size_bits;
+
+	if (!KEY_DIRTY(k))
+		return false;
+
+	stripe = KEY_START(k) >> dc->disk.stripe_size_bits;
+	while (1) {
+		if (atomic_read(dc->disk.stripe_sectors_dirty + stripe) !=
+		    stripe_size)
+			return false;
+
+		if (nr_sectors <= stripe_size)
+			return true;
+
+		nr_sectors -= stripe_size;
+		stripe++;
+	}
+}
+
 static void dirty_init(struct keybuf_key *w)
 {
 	struct dirty_io *io = w->private;
@@ -152,7 +177,22 @@ static void refill_dirty(struct closure *cl)
 		searched_from_start = true;
 	}
 
-	bch_refill_keybuf(dc->disk.c, buf, &end);
+	if (dc->partial_stripes_expensive) {
+		uint64_t i;
+
+		for (i = 0; i < dc->disk.nr_stripes; i++)
+			if (atomic_read(dc->disk.stripe_sectors_dirty + i) ==
+			    1 << dc->disk.stripe_size_bits)
+				goto full_stripes;
+
+		goto normal_refill;
+full_stripes:
+		bch_refill_keybuf(dc->disk.c, buf, &end,
+				  dirty_full_stripe_pred);
+	} else {
+normal_refill:
+		bch_refill_keybuf(dc->disk.c, buf, &end, dirty_pred);
+	}
 
 	if (bkey_cmp(&buf->last_scanned, &end) >= 0 && searched_from_start) {
 		/* Searched the entire btree  - delay awhile */
@@ -446,7 +486,7 @@ void bch_cached_dev_writeback_init(struct cached_dev *dc)
 	closure_init_unlocked(&dc->writeback);
 	init_rwsem(&dc->writeback_lock);
 
-	bch_keybuf_init(&dc->writeback_keys, dirty_pred);
+	bch_keybuf_init(&dc->writeback_keys);
 
 	dc->writeback_metadata		= true;
 	dc->writeback_running		= true;
diff --git a/drivers/md/bcache/writeback.h b/drivers/md/bcache/writeback.h
index 5ce9771df04..c91f61bb95b 100644
--- a/drivers/md/bcache/writeback.h
+++ b/drivers/md/bcache/writeback.h
@@ -1,6 +1,9 @@
 #ifndef _BCACHE_WRITEBACK_H
 #define _BCACHE_WRITEBACK_H
 
+#define CUTOFF_WRITEBACK	40
+#define CUTOFF_WRITEBACK_SYNC	70
+
 static inline uint64_t bcache_dev_sectors_dirty(struct bcache_device *d)
 {
 	uint64_t i, ret = 0;
@@ -11,6 +14,46 @@ static inline uint64_t bcache_dev_sectors_dirty(struct bcache_device *d)
 	return ret;
 }
 
+static inline bool bcache_dev_stripe_dirty(struct bcache_device *d,
+					   uint64_t offset,
+					   unsigned nr_sectors)
+{
+	uint64_t stripe = offset >> d->stripe_size_bits;
+
+	while (1) {
+		if (atomic_read(d->stripe_sectors_dirty + stripe))
+			return true;
+
+		if (nr_sectors <= 1 << d->stripe_size_bits)
+			return false;
+
+		nr_sectors -= 1 << d->stripe_size_bits;
+		stripe++;
+	}
+}
+
+static inline bool should_writeback(struct cached_dev *dc, struct bio *bio,
+				    unsigned cache_mode, bool would_skip)
+{
+	unsigned in_use = dc->disk.c->gc_stats.in_use;
+
+	if (cache_mode != CACHE_MODE_WRITEBACK ||
+	    atomic_read(&dc->disk.detaching) ||
+	    in_use > CUTOFF_WRITEBACK_SYNC)
+		return false;
+
+	if (dc->partial_stripes_expensive &&
+	    bcache_dev_stripe_dirty(&dc->disk, bio->bi_sector,
+				    bio_sectors(bio)))
+		return true;
+
+	if (would_skip)
+		return false;
+
+	return bio->bi_rw & REQ_SYNC ||
+		in_use <= CUTOFF_WRITEBACK;
+}
+
 void bcache_dev_sectors_dirty_add(struct cache_set *, unsigned, uint64_t, int);
 void bch_writeback_queue(struct cached_dev *);
 void bch_writeback_add(struct cached_dev *);